ABSTRACT
Previous studies have found that high-strength bolt axial force of many in-service bridges decreased by 10%–20% from the design bolt axial force. One of the reasons is that the bolts cannot be tensioned to the design axial force owing to various difficult construction conditions and errors. Appropriate control of the axial force while fastening the bolts is important. Contact methods using ultrasonic testing and eddy currents have been employed to detect bolt axial force. This paper proposes a non‐contact vision-based bolt axial force detection method using deep learning. We focus on the relative deformation feature of the bolt head inscriptions and train the neural network via regression method. The training results indicate that even when the image size is reduced to 448 pixels, it was able to obtain correct results for axial force detection for the type of bolt that was used to train the model.
